Cognition in Plants To appear in F. Baluška (Ed.) Plant - environment interactions: Behavioral perspective. Elsevier.

Page created by Ida Leonard
 
CONTINUE READING
To appear in F. Baluška (Ed.) Plant – environment interactions: Behavioral perspective. Elsevier.

Cognition in Plants

Paco Calvo Garzón1, Fred Keijzer2

1
  Department of Philosophy
University of Murcia, Spain
fjcalvo@um.es
2
 Faculty of Philosophy
University of Groningen, The Netherlands
F.A.Keijzer@rug.nl

13.1 Abstract
We discuss the possibility and the meaning of the claim that plants are cognitive from the
perspective of Embodied Cognition. In Embodied Cognition, the notion of cognition can be
interpreted very broad and applies to many free-moving creatures. In this chapter, we discuss
whether and, if so, how this approach applies to intelligence in plants. Building on work from
‘plant neurobiology,’ we discuss the differences in speed between plants and animals, similarities
between sensory-driven plant growth and animal memory, and the presence of offline behavior in
plants. In our view, these examples show that under a wide, embodied interpretation of cognition,
plants may well qualify as representatives.

13.2 Introduction
Plants exhibit much more complex and flexible adaptive behavior than most people would
expect (Trewavas 2003). With that being said, the very possibility of plant behavior as exhibiting
certain cognitive aspects strikes many people as outrageous, whether they are plant scientists
or not. In this chapter, we will discuss the possibility and meaning of claiming that plants
might be cognitive by turning to current discussions within the field of embodied and embedded
cognition (‘Embodied Cognition’, hereafter) and its links to biology. In this context, the notion of
cognition is drawn around perception-action phenomena and its nature applies much wider
than human-level thought.
    Embodied Cognition takes perception-action as its major focus, and within this embodied
perspective, most animal and even bacterial behavior can be considered cognitive in a limited
form. Embodied Cognition stresses the fact that free-moving creatures are not simple, hard-
wired reflex automatons but incorporate flexible and adaptive means to organize their
behavior in coherent ways. Even when it may go too far to ascribe a mind to such systems,
they deserve an acknowledgement of the intelligence involved in the things they do, and for
this reason, the notion of cognition seems appropriate.1 So far, the notion of cognition has not
been extended to plants. One reason is simply that most cognitive scientists, even those
involved in Embodied Cognition, are simply unaware of what plants can do. This is simply
remedied. However, a theoretically more important reason to exclude plants from the wider
cognitive domain sketched by Embodied Cognition seems to be the lack of the kind of

1
 It is thus important to differentiate here between this wide interpretation of cognition, which may apply in a
meaningful way far beyond the human case, and the notion of mind, which may well remain highly restrictive
and possibly limited to human beings.

                                                      1
sensorimotor organization that is the major focus of embodied cognition. We will focus on
this issue.
    In recent years, the issue of plant intelligence or cognition and even ‘plant neurobiology’ has
become a lively debate among plant scientists (e.g. Alpi et al. 2007; Barlow, 2008; Firn 2004;
Trewavas 2003, 2005; see also Calvo 2007, and references therein). So far, this debate has gone
unheeded within Embodied Cognition, nor has this debate itself relied on ideas from Embodied
Cognition. We think the discussions in the plant sciences can benefit from work in Embodied
Cognition, which has already done significant progress on clarifying the notion of cognition
and possible criteria for its use. At the same time, confronting Embodied Cognition with the
‘plant question’ is sure to raise important new inputs in the field of Embodied Cognition itself.
    In this chapter, we will discuss the question whether an extended reading of cognition, as
developed within Embodied Cognition might apply to plants. Within Embodied Cognition, the
notion of cognition – being based on perception and action – is used to make sense of a wide
range of behaviors exhibited by ‘simple’ animals, like nematodes or flies. The message is clearly
that we should set generalizing dismissive intuitions concerning such animals aside and go for a
more empirically informed approach. We believe that this open attitude is also beneficial to the
study of possible cognitive phenomena in plants.
    The chapter has the following structure. In section 3, we introduce the discussion on what
cognition is, and sketch the difficulties in clarifying this key notion for the cognitive sciences.
In section 4, we formulate a wide reading of cognition as derived within Embodied Cognition, also
taking in biological considerations. This wide reading is subsequently condensed into five
constraints on cognition, two of which explicitly highlight the need for an animal-like sensorimotor
organization. In section 5, using the work of Hans Jonas, we discuss why having a sensorimotor
organization is important for cognitive phenomena and whether plants may fulfill some of these
functions in alternative ways. In the next three sections, we discuss plant research findings from
which we conclude that, in this light, plants can fulfill these constraints to a significant degree.
We introduce the main tenets of plant neurobiology in section 6 and argue that the differences
in speed and form of plant behavior do not exclude a cognitive interpretation. In section 7, we
argue for the similarities between plant growth and animal memory. Section 8 deals with plant
control structures for forms of offline cognition. We conclude that these examples show that
plants can be considered to be minimally cognitive and to constitute an important domain for
cognitive studies.

13.3 What is cognition?
What is cognition? Although cognition is one of the core concepts in the behavioral and cognitive
sciences, there is no generally accepted answer. For example, in his classic book Cognitive
Psychology, Ulrich Neisser defined cognition as: “all processes by which the sensory input is
transformed, reduced, elaborated, stored, recovered, and used.” (1967, p.4) But this definition
seems to include many artifacts, like tape recorders, and organisms, like plants, that were not
intended to be labeled as cognitive. The classical cognitive sciences that grew up under the influence
of people like Neisser used a much more limited interpretation of cognition: not all forms or
information processing did suffice. The implicit extra constraint in this definition was that
cognition involves the kind of information processing that also occurs in human intelligence,
where it is described in terms like perception, planning, thinking and action.
    Because this field was intrinsically interested in the human mind and not in any other
topic, this implicit limitation worked quite well in practice. No one had to provide an answer
why tape recorders and plants were out, as these were not part of the topic studied. The notion
of cognition itself became widely interpreted as referring to the information processing
mechanisms that describe how the human mind operates. Human cognition became the default

                                                  2
interpretation of cognition and also the yardstick by which the cognitive abilities of other
animals were to be measured. From that perspective, cognition is tantamount to characteristically
human-like capabilities such as reasoning, problem-solving and symbolization. These processes
have to be present to a significant degree before one can speak of a bona fide cognizer (Gould
and Gould 1998). Internal, representation-handling processes are considered to be the source
of these particular thinking skills. It is a matter of painstaking research to establish whether, and if
so, which other creatures also exhibit these refined capabilities (see for example Heinrich 2000;
Smirnova et al. 2003). In the human-based interpretation of cognition, organisms whose behavior
does not unequivocally involve human-style-reasoning subsequently remain outside the cognitive
domain. Cognition is a scarce commodity in this view.
    Having an interest in human cognition, and not in the rest, is a way of carving up the
domain that leaves little motivation for articulating what falls outside this domain. In the cognitive
sciences, the behavior of nonhuman organisms receives comparatively little attention (for a notable
exception, see Bekoff et al. 2002). Such behavior is often argued to be predominantly composed
of inflexible, hard-wired reactions to environmental stimuli (e.g. Dennett 1984, 1996; Gould
and Gould 1998; Sterelny 2001). As a result such behaviors are not considered as being very
interesting from a cognitive perspective and only require closer attention when they come
nearer to the human forms of intelligence (see also Godfrey-Smith 1996, 2001; Shettleworth 1998).
Dennett (1984) and Hofstadter (1985), for example, talk about “sphexisms” in this context,
drawing the term from an anecdote in which a digger wasp of the Sphex genus was manipulated
so that it remained stuck in an iterative, automatic behavioral loop; endlessly repeating its own
inbuilt, behavioral responses.2 In other words, anthropocentric interpretations of cognition
depict a rough dichotomy between intelligent cognizers, and inflexible, mechanic-like organisms
merely capable of reflexive/instinctive behaviors.
    However, there are important practical and theoretical problems with such a general
dichotomy between human cognition and ‘non-cognitive’ inflexible, mechanistic behavior, as
well as with a human-based interpretation of cognition. First, this putative dichotomy does not
stroke with the underlying intricacies that make so-called ‘non-cognitive’ organisms tick, as it
simply fails to provide a realistic account of the behavioral complexities that can be found in
non-human organisms (Brooks 1999; Keijzer 2001; Roth and Wullimann 2001). When investigated
in their own right, the mechanisms and processes required to generate these presumably non-
cognitive behaviors are found to be very complex and extraordinarily difficult to replicate in
robots (Prescott et al. 1999).
    Second, the dichotomy does not cope with any differentiation within the so-called non-
cognitive organisms. There are huge gaps between the behavioral capabilities of, for instance,
nematodes and octopi, or between sharks and squirrel monkeys, all of which are - definitely to
plausibly - considered ‘non-cognitive’ from an anthropocentric perspective. The assumption
that the behavior of these ‘lower’ organisms is entirely composed of reflexes, instincts and/or
hardwired reactions does not help to articulate how such very different behavioral capabilities
come about.
    Third, when one turns to the basic processes of cognition, and leaves aside their anthropocentric
interpretation, it is clear that these processes, such as perception, memory, and action are dispersed
extremely widely across and even beyond the animal kingdom. It is now even plausibly
defended that these exemplar features of cognition are already present in invertebrates
(Carruthers 2004; Keijzer 2001), and prokaryotic bacteria (di Primio et al. 2000; Greenspan and
van Swinderen 2004; Lengeler et al. 2000; Müller et al. 2001).
    Fourth, a general dichotomy makes it more difficult to develop a gradualist and diversified
evolutionary account of how basic forms of cognition developed into different and more complex

2
    The story is actually a mere anecdote and not systematically corroborated by evidence (Keijzer, 2001)

                                                         3
ones. As it places the boundaries of cognition very high, say primates or, possibly, tool-users,
the relevance of preceding evolutionary stages of intelligent behavior is blurred. The differences
are stressed rather than possible continuities discovered.
     All these considerations were known for a long time but did not have much impact within the
cognitive sciences until the rise of Embodied Cognition. This new field of cognitive sciences
came to the fore in the late eighties and the nineties as a reaction to the strong focus on explicit,
reflective human thought. Hurley baptized this view ‘the sandwich model’ as it interpreted cognition
as a separate inner filling, wedged between perceptual input and behavioral output (Hurley, 1998).
Embodied Cognition, instead, stressed the ongoing, dynamic interactions between an agent and its
environment by means of perception and action. In this view, perception-action processes
became the starting point for intelligence, and complex human thought one of its offshoots.3
Perception-action – and the sensorimotor organization it requires – thus becomes the online
key feature of cognition, which is evolutionary expanded in higher-level cognizers to include
all sorts of offline processing that exhibits representational characteristics.
     One important consequence of Embodied Cognition was that the notion of cognition itself
started to shift away from its human anchor. Roboticists, for example, started to make insect-
like robots that had little to do with human cognition, but a lot with investigating a bottom up
interpretation of intelligence, and thus a kind of cognition. In this context, the link to the human
case became insufficient, leading for a call to formulate a mark of the cognitive (Adams and
Aizawa 2001). How such a mark could be given remains a problem, but there seems to be a
strong consensus that cognition involves processes such as perception, thinking, memory, and
action. As already stressed, it is useful to differentiate between mind and cognition here, even
though the two are often used intermingled within the cognitive sciences. While ‘mind’ may
or may not remain closely tied to humans and in particular to consciousness, ‘cognition’ need not
be intrinsically connected to either humans or consciousness. Thus while ‘cognition’ is a
concept that can be stretched beyond its original use, this can be done without additional claims
that all cognitive systems are to be seen as mindful in a way that is similar to human minds.
     We think that the notion of cognition ought and can be developed in new ways to fill in
the gap between the mindful and the mindless, and turn it into a gradient from human
intelligence to inanimate nature. Godfrey-Smith (2001, p. 234) argued that cognition in the
first place evolved to enable organisms to control their own behavior, allowing them to cope
with environmental complexity. In his view, cognition “shades off” into basic biological
processes such as metabolism. We believe that a proper interpretation of cognition should aim
to become more specific about this “shading off”, and allow for a better differentiation within
the wide array of cognitive capabilities that are found in nature. In the next section, we will
develop the outlines of a wide interpretation of cognition, as it can be derived from ongoing
theoretical developments within embodied and biological views on cognition.

13.4 A biological and embodied perspective on cognition
Since Embodied Cognition provides the intrinsic connection between humans and cognition,
work has begun to provide a more systematic account of cognition. Taking in artificial systems
like robots makes this task very difficult and possibly impossible. For example, it is hard to
make sense of the question whether a bacterium is as intelligent as a washing machine (Firn, 2004),
because it is utterly unclear on which aspects of both one should focus. However, restricting
oneself to naturally occurring, biological systems provides more grip (e.g. Barandiaran 2008;
Keijzer 2001, 2003, 2006; Lyon 2006a, 2006b; Moreno and Etxeberria 2005; Moreno et al.
1997; Van Duijn et al. 2006). The main idea here is that cognition is (or originated as) a
3
 For introductions and overviews of Embodied Cognition, see e.g. Calvo and Gomila (2008), Clark (1997),
Pfeifer and Scheier (1999), or Varela, Thompson, and Rosch (1991).

                                                     4
biological phenomenon and is used to manipulate the environment in ways that systematically
benefit the living organization that exhibits these cognitive aspects. In this context, the question
concerning minimal cognition is important (Beer, 2003): What is the minimal biological system
to which the notion of cognition applies?
         We will disregard claims that make life itself a form of cognition, but it can be argued
that bacteria already provide examples of cognition (di Primio et al. 2000; Lengeler et al. 2000;
Lyon 2006a; Van Duijn et al. 2006). Consider chemotaxis in E. coli. These free-moving
bacteria use flagella to move around and can travel up or down gradients of several substances
that they can ingest or need to avoid. All the basic ingredients for a minimal form of cognition
are already present here, and discussing them will provide a suitable framework when it comes
to judging the possibility of plant cognition.
     As a living organism, E. coli involves a metabolic biochemical organization, which provides
the fundamental energetic and constitutive processes of the bacterium, as it does for all living
organisms. To stay alive, the bacterium needs substances that it can incorporate into its structures
and it needs energy to drive these biochemical processes. This metabolic organization provides
a basic form of normativity (Bickhard 2008) – differentiating ‘bad for me from good for me’ –
and can itself have adaptive characteristics. For example, a well-known form of metabolic
adaptation is the ‘lac operon’ system, which regulates the metabolism of lactose in E. coli.
This cluster of genes is normally dormant, because the bacterium predominantly metabolizes
glucose. However, when the bacterium detects that glucose levels are very low and lactose is
abundantly present in the environment, the lac operon system becomes dis-inhibited, subsequently
allowing the transcription and expression of genes that enable lactose metabolism (Todar 2004).
This form of metabolic adaptation is induced by environmental conditions, but still is a part of
the organism’s metabolic organization. The process consists of a change in the set of chemical
reactions that together constitute the bacterium’s metabolism.
     Chemotaxis, on the other hand, is a different kind of process. It is itself not constituted by
chemical reactions, but by physical changes in the position of the bacterium with respect to its
environment. In other words, the bacterium interacts with its environment at the larger, physical
scale of the dispersion of metabolically relevant substances. There are potential metabolic benefits
to this interaction, but this manipulation of the environment—moving towards a food source,
for example—is itself not a metabolic process. With respect to metabolism, chemotaxis is a
second-order process, which is relevant for changing metabolic opportunities, and in this way
expanding the adaptive opportunities of organisms to a considerable degree.
     Manipulating the environment at larger scales to enable or enhance metabolic functioning
is a very general biological strategy. Sponges pumping water through their bodies, plants growing
leaves oriented toward the light, lions stalking their prey, and even humans discussing which
restaurant to go to can be considered as examples. Why would all of these activities be cognitive? In
the Embodied Cognition literature there is a clear intuitive cut off point, which limits cognition
to systems showing a form of sensorimotor coordination. Cognition applies to free-moving agents,
capable of reversible movements and perception. In this view, bacterial chemotaxis is a possible
example of minimal cognition as it uses sensorimotor coordination to expand metabolic forms of
adaptation and, in this way, provides a basic example of an organization that is also present in
human cognition. Plants, fungi and sessile animals, however, would be left out of the cognitive
domain insofar as they (seem to) lack this additional requirement.
     Setting up an adequate sensorimotor organization requires a particular physical embodiment of
an organism, be it bacterium or monkey. For bacteria, this comes in the form of specific chemical
receptors such as methyl-accepting proteins, and actuators such as flagella or pili that enable
the bacterium to move about (Berg 2000). It also involves a control system that enables the
organism to initiate motion and use the perceptual feedback it generates to guide this motion.
Within Embodied Cognition, it is customary to differentiate between online processing that is

                                                 5
under direct perceptual control, and offline processing which is to some extent decoupled from
immediate perception-action. Online processing is cast as more basic, while offline processing
is thought to be involved in more complex cognitive tasks, like those relating to memory or planning.
     Importantly, E. coli already shows some offline capacities. It does not measure a concentration
gradient by calculating the simultaneous difference between different sensors at different
body-parts. Instead, E. coli, and many other bacteria, uses a TCST-system which acts as a
memory and inner connection between sensors and effectors in a way that is functionally
similar to the nervous system in multicellular animals. The TCST-system allows it to take
sequential measurements of the substance concentration. When the bacterium happens to
travel down the gradient, according to this memory, it tends to change directions randomly,
and when it travels up the gradient, E. coli tends to maintain its course. The net result is a
systematic form of chemotaxis.
      Finally, in this biological and embodied perspective on cognition, the sensorimotor organization
does not consist of simple and hard-wired behavior. All animal behavior results from a complex
non-linear dynamical process involving a nervous system, a body and an environment that interact
with one another on a continuous basis (Beer 1995, 2000; Keijzer 2001). Even seemingly
repetitive and automatic behavior is still the result of complex non-linear processes, which show up
when the larger behavioral repertoire of the organisms is taken into account. This also applies
to our example of E. coli. We want to stress that the system’s chemotaxis is actually much
more complex than suggested so far, as it also adapts to the absolute concentration level of the
substances, works in conjunction with sensitivity to several other substances, and is in itself
part of a very flexible organization for behavioral control (Lyon 2006a). Merleau Ponty (1963)
provided a clear and detailed analysis of how the seeming simplicity of environmentally
controlled stimulus-response behavior results very easily as a methodological artifact when
one tries to isolate a specific aspect of behavior from ordinary behavior of an animal.
     To summarize, casting biological forms of sensorimotor coordination as the minimal form
of cognition provides a clear and transparent starting point for thinking about cognition. In our
view, by framing the issue of the nature of cognition from within the field of Embodied Cognition
the question concerning plant intelligence can be suitably addressed. The problem to be solved
first, however, is whether plants are not to be excluded from the cognitive domain as envisioned
within Embodied Cognition.

13. 5 Embodied Cognition and plants
An embodied and biological perspective introduces a more open-ended interpretation of cognition
and intelligence. The upshot of this enterprise can be summarized in five different constraints
that apply to cognition as conceived in this way:
    1. Metabolism provides a basic form of biochemical normativity for cognition.
    2. Cognition proper (initially) consists of exploiting the spatio-temporal dispersal
        characteristics of metabolically relevant environmental features by movement of the
        organism.
    3. This movement takes the form of various sensorimotor organizations.
    4. A basic, online, sensorimotor organization can be expanded by using offline control
        structures that can, but need not, involve a nervous system.
    5. Such a sensorimotor-based cognitive organization is a globally organized cohering
        unit, not a collection of individual stimulus-response relations.
The question to ask now is whether and, if so, how does this set of constraints apply to plants?
To begin with, we must stress that the application of the first two constraints is not disputed. Plants
metabolize, of course, giving them a basic motivational setup for doing things. There is no
doubt either that plants manipulate their environment in a second-order way such that their

                                                  6
metabolic functions profit from this manipulation. Growing roots downward and light-catching
parts upward suffice in this respect (Keijzer 2001).
     The real rub for minimal cognition in plants comes from constraints 3 to 5. Constraint 3
imposes being free-moving, having a sensorimotor organization, as a requirement for cognition.
It is here that the option of plant cognition seems highly problematical within an embodied
perspective. However, adhering to being free-moving as an intuitive criterion is unsatisfactory.
Within the biological domain, free-moving organisms may stand out as potentially intelligent
beings, but why should we trust these intuitions? The question should rather be: Why would we
consider being free-moving so important?
     Up to now, plants have not received much attention within Embodied Cognition. Most of
the field worked with a default assumption that intelligence was at a minimum an animal thing
that was best caught in studies with free-moving agents such as robots, while excluding
sessile plants. However, Hans Jonas (1966, 1968), now receiving renewed attention as an
important thinker on the connections between biology and mind (Barandiaran 2008; Di Paolo
2005; Keijzer 2006), did take plants into consideration. Jonas tried to articulate the differences
between plants, animals and humans in a way that highlights the relevance of being free-moving. In
his view, the capacity for free movement is a key feature that is required for the development of
intelligence as exhibited by animals, and a precondition for the evolution of human thought. Jonas’
work can be used to argue that there exist fundamental differences between intelligence in plants
and animals (e.g. Barandiaran 2008). At the same time, by clarifying in what way being free-moving
is so important, he provides a clearer target to be challenged on empirical and theoretical
grounds. Plants may very well fulfill these constraints, without moving about like animals do.
     Jonas provides an analysis why motility, and in its wake sensing and emotion are key features
when it comes to cognition. 4 He argues that animal motion is more than an intensified case of
vegetative motion from which it differs in a number of physical respects: “in speed and spatial
scale; in being occasional instead of continual; variable instead of predefined; reversible instead
of irreversible.” (Jonas 1968, p. 248). These criteria are important to make the difference
between being free-moving – which plants are not, generally speaking – and having self-induced
motility – which is present in plants. Jonas uses these physical differences as a foundation for
a further argument that animal motion leads to a principled new way of coordination with the
geometry of environmental space (1966, see also Barandiaran, 2008). Jonas makes the point as
follows:
         Now it is the main characteristic of animal evolution as distinct from plant life that
         space, as the dimension of dependence, is progressively transformed into a dimension
         of freedom by the parallel evolution of these two powers: to move about, and to
         perceive at a distance. (Jonas 1966, p. 100).

    In his view, only by free moving and perceiving at a distance, most notably by vision, is
“space really disclosed to life.” The key issue is that (aspects of) the global spatial structure
of the environment must become a feature that is present and accessible for an organism. A
fly, for example, is able to orient itself within its environment and home in on places with
sweet stuff, while avoiding swapping hands. Animals are sensitive to the spatial layout of the
environment, for example in the form of patterns on a sensory surface like the retina or the
skin, and their behavior is globally organized as a unit in relation to this layout. Barandiaran
uses the nice phrase of being sensitive to the “geometric space where objects can be freely
explored” (2008, p.198). Such sensitivity comes in different grades, as the fly will not be
sensitive to the highly relevant fact for me that it is trying to land on my child’s birthday cake.

4
 For Jonas (1966), motility and perception are also intrinsically linked to emotion and the presence of an inner,
phenomenal dimension. We will not discuss these further complexities here.

                                                       7
However, in both cases, the fly and me are sensitive or ‘aware,’ that of the environment as a
spatially and temporally extended structure in which we can act.
    In contrast, plants are presumed to act on local stimuli, which may guide its behavior in
globally appropriate ways, but without being directly sensitive to the spatial patterning of the
environment. Thus, plants may grow their roots systematically downward strictly based on the
locally available perception of gravity in every root. In this way, they can exploit this geometric
structure without being sensitive as a single unit to geometric space as induced by free motility.
The issue is the extent to which plants are sensitive to and acting on such global spatial structure
of the environment. Or do they get by on the basis of a multitude of local interactions or
decision making processes? Thus, the important challenge that Jonas brings forward here for
plant cognition is that only free motility seems to bring forth an independent world – a
geometric space – in which an agent can act.
    Importantly, Jonas shifts the issue from a general unspecific commitment to a sensorimotor
organization that plants just do not have, to different, more specific demands that require
empirical data to settle. In line with constraint 3, Jonas changes the issue from having an
animal-like sensorimotor organization to motility and possible differences in the speed,
variability and reversibility of motility. As we will see, recent developments in plant science
provide good empirical reasons to downplay the differences between being free-moving and
self-induced motility, as imposed by constraint 3. Similarly, sensitivity to the geometric lay
out of the environment as stressed by Jonas, may be something that plants are quite capable of
without us knowing it. Thus, plants could also very well fulfill constraint 5, making this also
an empirical issue rather than a theoretical one.
    Constraint 4 stresses the importance of offline control as a way to expand the options of
an online operating control structure. Offline control is often considered as a very important
sign of cognition. Offline control allows an organism to dissociate its behavior from the
immediately impinging stimuli and to act in ways that are guided by forms of knowledge. It is
ironic then that these more cognitive offline aspects can be comparatively easy established in
plants, when compared to the motility issue.
    In the following three sections, we will turn to plant science and discuss developments and
examples that make it plausible that plants can fulfill constraints 1 to 5 to some degree, and thus
can be deemed cognitive in a sense that is regularly used within Embodied Cognition. In particular,
the development of plant neurobiology has been important in establishing many details of plant
intelligence.

13.6 Plant neurobiology: Intelligence can take different forms and speeds
Plant neurobiology (Baluška et al. 2006) has emerged in the last few years out of the integration
of results from areas of research such as plant electrophysiology, cell biology, molecular biology,
and ecology. The difference between plant neurobiology and other more basic disciplines
resides in the target of these interdisciplinary efforts. Plant neurobiology adds up a scientific
understanding of the integration of plant sensing and responding. The target is the scientific
understanding of how metabolism and growth can be regulated by the endogenous integration
and processing of information. More specifically, plant neurobiology lays the stress on integrated
signaling and electrophysiological properties of plant networks of cells. As Baluška et al. (2006)
point out:
      “Each root apex is proposed to harbor brain-like units of the nervous system of plants.
      The number of root apices in the plant body is high, and all “brain units” are interconnected
      via vascular strands (plant neurons) with their polarly-transported auxin (plant neurotransmitter),
      to form a serial (parallel) neuronal system of plants.” (p. 28).

                                                   8
The working hypothesis of plant neurobiology is that the integration and transmission of information
at the plant level involves neuro-like processes such as action potentials, long-distance electrical
signaling, and vesicle-mediated transport of (neurotransmitter-like) auxin (Brenner et al. 2006).
    Interestingly, from the point of view of the study of plants qua information-processing
systems, the issue of plant intelligence is less mysterious, or at least subject to investigation in
the very same way that cognitive neuroscience deals with in its respective target domain. As recent
research in plant neurobiology shows, plants are not passive systems that build up photosynthate.
Plants exhibit sophisticated forms of behavior, being able to assess current data that can suppose
an advantage at a later stage. Roots, for instance, exhibit patterns of growth that depend upon
future acquisition of minerals and water. Plants are indeed sensitive to a minimum of 15 biotic
and abiotic signals, among which are not only water, light, minerals, or gravity, but also soil structure,
neighbor competition, herbivory, allelopathy, and wind, to name but a few (Trewavas 2000).
Likewise, plant roots can, for instance, sense volume, discriminate self from alien non-self roots,
and allow for phenotypic root reordering as a function of competition for nutrients. Thus,
current evidence easily shows that plant behavior is not at all predefined, as Jonas claimed, but
highly variable.5
    The commonalities in variability of the behavioral repertoires of animals and plants can be
strengthened when their different functional and architectural setups are taken into account.
The first thing to note concerning plants is that the architectural constraints on cognition may
be pretty minimal. When Darwin took the Venus’ flytrap as a clear example of a plant with
“animal” features (1875), he was not searching for brains in plant’s root tips, even though
they clearly exhibit sophisticated forms of computation. Darwin (1880) concluded: “…the tip
of the root acts like the brain of one of the lower animals, the brain being seated within the
anterior end of the body receiving impressions from the sense organs and directing the several
movements” (p. 573). As Darwin observed, plant forms of behavior can be highly sophisticated.
    Second, although perception, memory and action are capacities that can be present in both
animals and plants, they take different forms. Animals, insofar as they are heterotrophic
organisms that require organic foodstuff to survive, exploit a number of mobility-related
competencies in order to navigate in complex and contingent environments (Neumann 2006).
Animals also appear to be better fitted to escape from predators or harmful environments.
Plants, by contrast, do not require contractile muscles for fast responses to environmental
contingencies. Insofar as plants are autotrophic organisms, they operate in slower time scales
since inorganic substrates can be synthesized into organic compounds while remaining stationary.
This also means that the computational solutions found by both plants and animals can diverge
even when they exhibit similar complex functions.
       Someone may nevertheless argue that at the level of the architecture a borderline should be
drawn between the two kingdoms. Animal neural networks allow for intelligent behavior courtesy
of a hardware arranged in parallel (Rumelhart et al. 1986), as opposed to a serial one. However, in
our view, it would be a mistake to demand the same type of computational architecture in plants
and animals to accommodate constraints 3 and 4. Different architectures can serve to approximate
the same function. Non-linearly separable functions may be computed, for instance, by inserting
a layer of processing units in between the sensory and the motor layers. In this way, the metric
relations of sensory similarity get recoded in terms of functional relations that allow for the
approximation of the sensorimotor function, as the logical operation of exclusive disjunction
(aka the XOR problem) has taught cognitive scientists. But there are many different ways in
which a non-linearly separable function can be computed, and inserting a layer in between is
not a prerequisite. A two-layer network can equally approximate the function, simply by choosing
the right sort of activation function for the problem at hand, such as the cosine or sine transfer
5
    For more sophisticated plant competencies, see Trewavas, 2005.

                                                        9
functions for solving the XOR problem (Rosen et al. 1990). The bearing upon our current
concern is thus that in order for plants to exhibit sophisticated forms of computation, a parallel
operating neural network may not be necessary.
    Thus, differences in speed and architecture have fostered the idea that plant behavior, compared to
animal behavior, is strictly determinate and invariant under a variety of conditions. But animal and
plant avoidance responses are both graded as a function of the stimulus strength and involve
equally modifications in cellular morphology. Plant behavioral invariance is in the eye of the
anthropocentric - or heterotrophicentric, we may say - observer. But, once we approach the
study of (minimal) cognition non-anthropocentrically, perception, memory, and action are
common currency across phylogeny.
      Summing up, differences in speed and form do not serve to exclude a cognitive interpretation
of plant behavior. In our view, the issue is not whether plants can move about, acting intelligently
in this way, but rather whether plants, being autotrophic organisms, integrate information, have
memory, can make decisions, in such a way that their adaptive coupling to their environment
can be called ‘cognitive’. Light, gravity, moisture, and touch are signals that plants integrate
and respond to in complex non-linear ways. Roots make decisions in particular contexts as to
which type of signal(s) to honor (Li & Zhang, 2008). Furthermore, phenotypic plasticity in ever-
changing environments requires the exploitation of memory resources. Plants integrate exogenous
and endogenous information channels in an attempt to phenotypically adapt to environmental
contingencies; a sophisticated form of competency that, we believe, can be paralleled to animals’
predictive behavior. In the remainder of this chapter, we are further highlighting the similarities
between plant growth and animal memory, and try to show why the behavior of plants, as they are
coupled to their environment, allows for a cognitive interpretation of their adaptive responses.

13. 7 Similarities between growth and memory
Take carnivorous D. muscipula and A. vesiculosa. In the case of D. muscipula, an action potential
(AP) is generated whenever an upper trap hair is bent. Crucially, a single stimulation of the
hair does not trigger the closing of the trap. For the trap to close it is necessary a second AP
that takes place only when another hair is bent within 40 secs after the first AP has been
generated (Baluška et al. 2006). This setup forms a basic form of memory, similar to the TCST
system in bacteria (di Primio et al. 2000) as well as basic forms of animal memory.
    Or consider plant’s avoidance responses in relation to drought (Trewavas 2003). Drought
avoidance behavior results in a reduction in the rate of cell growth that involves, on the one hand,
changes in cytosolic Ca2+, [Ca2+]i, and in other second messengers, and, on the other hand,
phosphorylation changes in ATPases and associated ion channels related to turgor (Palmgren
2001). Trewavas (2003) compares drought avoidance responses in plants with the avoidance
behavior of Aplysia. The pattern of avoidance of this marine slug involves a form of short-term
memory whose mechanism includes Ca2+ and, in addition to the second messengers, cyclic
nucleotides and several protein kinases that operate as a temporary memory (Greengard 2001) by
phosphorylating ion channels.
    Phenotypic plasticity underlies the capability to memorize. Genesis of dendrites delivers
the goods in animal brains by effectively altering the architectural features of the network.
Different patterns of connectivity permit the network to acquire new functions. Plant networks
cannot exploit phenotypic plasticity with the computational resources of animal networks. For
one thing, plants’ lack neuron-like cells as computational building blocks where new dendrites
allow for new functional profiles overall. However, by contrast, plant cell divisions continue
at any stage of development. This means that the way to acquire different functionalities throughout
the life of the plant will need to differ in architectural terms somehow. Fortunately, as we saw
earlier, this need not be an insurmountable hurdle since intelligence is not to be univocally

                                                  10
linked to a specific type of architecture. Trewavas (2003) points out where the architectural
divergence may lie: “Just as the process of learning in a brain could be represented as a time
series, a set of snapshots of developing brain connections, in plants, each snapshot may possibly
be represented by developing plasmodesmatal connections or equally, successive new tissues.
So, instead of changing dendrite connections, plants form new networks by creating new tissues,
a series of developing brains as it were” (p. 14). In this view, modifying patterns of connectivity is
not what allows the plant network to remain competent. Rather, as new tissue accumulates,
new networks with different computational resources are stored on top of each other. But note
that newer tissue networks do not replace former ones. Rather, we have a succession of
operative serial networks, obtained as cells keep dividing throughout the life of the plant.
       The key issue in settling upon a non-biased approach to intelligence, as Trewavas (2003)
points out, resides in the “delays in the transfer of information between the sensory system
and the motor tissues acting upon the signals” (p. 1). But note that these delays cancel out a
rendering of plant forms of memory with some class of developmental progression, where previous
cellular states determine linearly future outcomes (Firn 2004). As Bose and Karmakar (2003)
point out, animal neuronal networks and plant calcium signaling systems are not that different
in terms of non-linearities. In the case of plants, non-linearities obtain by means of the succession
of signaling networks. Chakrabarti and Dutta (2003) have put forward an electrical network
that models plant calcium signaling systems and approximates Boolean functions, including XOR.
Open/closed ion channels play the role of neurons in networks of plants. As calcium ions are
released, diffusion across nearby channels gives rise to further calcium release, ultimately
giving rise to a calcium wave that flows throughout the whole network. Such a modeling of
calcium signaling networks, Bose and Karmakar (2003) argue, illustrates how memory mechanisms
can be implemented in plants. The dynamics of the calcium wave are governed by non-linear
equations, opening up the possibility to integrate and compute all incoming data (Trewavas,
2002) in a way that functionally resembles animal-based computation.
     If animal memory is characterized by processes that are not linear, the same goes for plants.
For a case in point of transfer delay in sensorimotor terms, consider the orchestrated behavior
of roots and shoots. Here, there is genuine emergent behavior that cannot be accounted for by
summing the local behavior of roots and shoots. Put bluntly, root-shoot interactions can only
be understood synergistically (Corning 2003). Or, to put it in computational terms, the XOR
function exemplifies the basis of complex behavior insofar as it exploits the delays in the transfer
of information between sensory and motor processing tissues, as the metric relations in sensory
space are altered in order for the problem to be approximated. We thus have the means to
overcome sensorimotor organization as a restriction (constraint 3, above). In what follows, we
illustrate from a behavioral perspective how plants have offline capabilities (constraint 4) that
allow them to do more than reacting to immediate stimuli.

13.8 Offline cognition: Leaf heliotropism
Someone may argue that the issue is not whether the function to be computed is linearly or
non-linearly separable, embodied or not, but rather whether the recoding of the metric relations that
obtain as a result of the approximation of the function can be interpreted in representational terms.
In other words, the issue is whether plants, however complex calcium wave networks happen
to be, manipulate representational states or not. Take the case of the stilt palm. In order to avoid
competition for light, the stilt palm (Allen 1977) “walks” away from shade and into sunlight.
The stilt palm grows new roots in the direction of sunlight, allowing older roots to die. Although
Trewavas (2003) interprets this as an intentional form of light-foraging behavior, it is usually
assumed that de-coupled, offline modeling tasks are what distinguish sophisticated forms of

                                                 11
behavior from merely tropistic online routines. Put bluntly, the sine qua non of representation-
based competency is offline adaptive behavior (Clark, 1997).
     Generally speaking, tropisms involve a directional change such as growth or movement in
response to a given environmental stimulation (Stanton and Galen 1993). Nevertheless, plant
tropisms can vary substantially as a function of the type of stimulus that the plant is responsive
to, and the part of the plant that responds to the stimulation. Complex, although still online,
tropistic reactions are frequent. Roots that manifest gravitropism stop developing downwards
as they encounter a physical obstacle, and grow horizontally instead. However, they are able
to assess online the state of affairs, and periodically try to move downwards, remaining horizontal
if unable to respond gravitropically (see Massa and Gilroy 2003). Likewise, we may think of
thigmotropism, hydrotropism, or thermotropism, as other sophisticated, although possibly, still online,
tropistic responses.
     But some plant tropisms do constitute a form of offline anticipatory behavior. In relation to
light-related tropisms, we can distinguish phototropisms, which may represent directed responses
to a static light source, from heliotropism, a more complex response that involves a correlated
response to changes in sunlight orientation as the day changes from sunrise to sunset. We may
also differentiate between flower heliotropism and leaf heliotropism. In the case of flower
heliotropism, no “memory” mechanisms seem to be required for flowers to keep track of the
position of the sun. Unless flowers are exposed to light in the morning they will fail to reorient to
sunrise, remaining in a random orientation throughout the night.
     But offline nocturnal reorientation by plant leaves represents a qualitative change with
regard to stimulus controlled online behavior. Leaf laminas of Lavatera cretica can, not only
anticipate the direction of the sunrise, but also allow for this anticipatory behavior to be retained
for a number of days in the absence of solar-tracking. That is, the laminas reorient during the
night and keep facing the direction of the sunrise even after a few days without tracking the
sun, and without sensing the position of sunset. Schwartz and Koller (1986) report a series of
experiments that clearly shows that this is a complex offline response. Three groups of plants
were taken at sunset to 3 different cabinets. One was kept in darkness; another had illumination in
day-light hours, and another one was kept illuminated vertically throughout the experiment.
The first day, plants in cabinets 1 and 2 (darkness and day-light hours conditions) could anticipate
sunrise and leaves were oriented towards that direction. Plants in cabinet 3, on the other hand,
were horizontal, with laminas facing upwards, as they had been kept under constant vertical
artificial illumination. The nocturnal reorientation behavior exhibited by plants in cabinets 1
and 2 lasted for as long as 3 days under the same experimental conditions, that is, in the absence
of day-time solar-tracking. In their study, the explanation of nocturnal reorientation cannot be
the sunrise of the day before, since plants were prevented from tracking the sun for 3-4 days.
     The explanation of nocturnal reorientation involves the internal modeling of environmental
rhythms. Circadian clocks allow for time-estimation by synchronizing endogenously generated
activity with exogenous cyclic periods such as day-night planetary patterns. Circadian clocks
can mimic biological rhythms on a 24-hour cycle, and this explains nocturnal reorientation in
plants for up to 4 days in the absence of sunrise stimulation. Plant genetics points towards
underlying shared molecular components that explain day-length estimations and the operation of
light receptors (see Pruitt et al. 2003, and references therein). In the case of time-estimation,
recent research in genomics has unearthed the molecular mechanisms of plant overt behavior,
with the result that both plants and animals draw on the very same molecular networks
(Cashmore 2003) in their adaptive exploitation of circadian clocks. One single level, thus,
explains the origins of the anticipatory behavior as exemplified by circadian clocks.
     As we saw earlier, the type of restrictions imposed by constraints 3 and 4 serve to stress
the importance of having a sensorimotor organization, on the one hand, and improving the offline
and information processing capacities of the system, on the other. Offline plant behavior may thus

                                                  12
be considered minimally cognitive insofar as information is processed flexibly and adaptively
in accordance with the aforementioned restraints. Constraint 5, nonetheless, remains an open
empirical question, although there are reasons to be optimistic, as we point out in the closing section.

13.9 Concluding remarks
The notion of cognition is very much in flux. Within Embodied Cognition the meaning of
cognition has been changed from human thought to a broader interpretation centered on
perception and action. In this perspective, free-moving animals are part of the cognitive domain
but sessile plants, without clear perception and action features and a dedicated sensorimotor
organization, form a more difficult case. Given that plants are definitely capable of many forms
of complex behavior, they provide an interesting domain for both questioning and clarifying
Embodied Cognition’s focus on free-moving agents. At the same time, Embodied Cognition’s
wide reading of cognition can reinforce the study of intelligence in plants, as it now occurs
within the field of plant neurobiology.
     We discussed how Embodied Cognition interpreted cognitive phenomena, and summarized
this view in terms of five constraints on cognition (Section 5). Two of these constraints (3 and 5)
emphasize being free-moving as a definite requirement for minimal forms of cognition. This
seems to contradict the possibility of plant cognition. We used an analysis of Hans Jonas to
articulate the requirements for being considered ‘free-moving’ and to assess possible reasons for
its importance. Subsequently, we argued that these two constraints should be readjusted, allowing
the possibility that plants fulfill these constraints in ways that differ from those in free-moving
creatures. We then discussed work in plants sciences to see whether it is plausible that plants can
fulfill the five constraints formulated above.
     Plants fulfill the first two of these constraints: having a metabolic organization that provides a
normative context, and exploiting environmental spatio-temporal structure to enhance metabolic
functioning. Our discussion centered on evidence for constraint 3 (sensorimotor organization),
4 (offline control), and 5 (acting as a global unit). Interestingly, while it is relatively easy to
establish the presence of offline control in plants, which is generally considered the key notion of
cognition, the evidence concerning the role and organization of motility and perception in plants
remains more equivocal and difficult to interpret.
     There is a large gap between facts and possible interpretations here. New developments in
plant neurobiology have made it clear that plants are capable of much more complex behavior
than many of us tended to think, including Jonas. We have argued that it is plausible that plants
can circumvent the requirement for an animal sensorimotor organization. They are capable of
organizing their behavior in ways that are different but still highly complex and adaptive, making
this behavior cognitive in the general and minimal sense used within Embodied Cognition. It
remains to be seen to what extent plants can really fulfill constraint 5. Plants integrate information
simultaneously provided by many different sensors on a variety of parameters in real time and
it may very well be that this allows them their own equivalent of the sensitivity provided by
animal sensory surfaces. We expect that plant neurobiology will clarify this issue further in
the near feature given its emphasis and focus on the integration of information present in different
parts of the plant.
     Concluding, we hope to have shown that, in a number of specific issues relating to cognition,
animals and plants do not fundamentally differ, and plants are cognitive in a minimal, embodied
sense that also applies to many animals and even bacteria. The scientific target in both cases is
to understand the continuous interplay of animals and plants in relation to the environmental
contingencies that impinge upon them. Plant cognition is in this view not a contradiction at all
but an empirical issue that requires much more attention, not only from plant scientists but
also from cognitive scientists, more generally.

                                                  13
13.10 Acknowledgments.
Fred Keijzer wants to thank Pamela Lyon, Marc van Duijn and Daan Franken for their helpful
comments and discussion. Preparation of this chapter was supported in part by DGICYT
Project HUM2006-11603-C02-01 (Spanish Ministry of Science and Education and Feder
Funds) to Paco Calvo Garzón.

13.11 References
Adams F, Aizawa K (2001) The bounds of cognition. Philos Psychol 14:43-64
Allen PH (1977) The rain forests of Golfo Dulce. Stanford: Stanford University Press.
Alpi A et al (2007) Plant neurobiology: no brain, no gain? Trends Plant Sci 12:135-136
Baluška F, Mancuso S, Volkmann D, Barlow P (2004) Root apices as plant command centres:
   the unique ‘brain-like’ status of the root apex transition zone. Biologia 59:9-17
Baluška F, Mancuso S, Volkmann D (eds) (2006) Communication in Plants: Neuronal Aspects
   of Plant Life: Springer-Verlag
Barandiaran X (2008) Mental life: A naturalized approach to the autonomy of cognitive agents.
   PhD thesis. University of the Basque Country, San Sebastian, Spain
Barlow PW (2008) Reflections on ‘plant neurobiology’ BioSystems 99:132-147
Beer R (1995) A dynamical systems perspective on agent-environment interaction. Artific Int
   72:173-215
Beer RD (2000) Dynamical approaches to cognitive science. Trends Cogn Sci 4:91-99
Beer RD (2003) The dynamics of active categorical perception in an evolved model agent.
   Adapt Behav 11:209-243
Bekoff M, Allen C, Burghardt GM (eds) (2002) The Cognitive Animal: Empirical and
   Theoretical Perspectives on Animal Cognition, Cambridge, Mass.: MIT Press.
Berg HC (2000) Motile behavior of bacteria. Phys Today 53:24
Bickhard M (2008) Is embodiment necessary? In: Calvo P, Gomila T (eds), Handbook of
      Cognitive Science: An Embodied Approach, Elsevier: 29-40
Bose I, Karmakar R (2003) Simple models of plant learning and memory. Physica Scripta
        T106:9-12.
Brenner ED, Stahlberg R, Mancuso S, Vivanco J, Baluška F, Van Volkenburgh E (2006) Plant
        neurobiology: an integrated view of plant signaling. Trends Plant Sci 11:413-419.
Brenner ED, Stahlberg R, Mancuso S, Baluška F, Van Volkenburgh E. (2007) Plant neurobiology:
        The gain is more than the name. Trends Plant Sci
Brooks R (1999) Cambrian Intelligence. Cambridge: The MIT Press.
Calvo Garzón P (2007) The quest for cognition in plant neurobiology. Plant Signal Behav
        2:208-211.
Calvo Garzón P, Gomila T (eds) (2008) Handbook of Cognitive Science: An Embodied
      Approach, Elsevier.
Carruthers P (2004) On being simple minded. Am Philos Quart 41:205-220.
Cashmore AR (2003) Cryptochromes: enabling plants and animals to determine circadian time.
        Cell 114:537-543.
Chakrabarti BK, Dutta O (2003) An electrical network model of plant intelligence. Ind J Phys
        77A:549-551
Clark A (1997) Being there: putting brain, body, and world together again. Cambridge: The
        MIT Press
Corning P (2003) Natures magic. Synergy in evolution and the fate of humankind.
   Cambridge: Cambridge University Press
Darwin C (1875) Insectivorous plants. John Murray
Darwin C (1880) The Power of Movements in Plants. John Murray

                                             14
You can also read